Governing AI-Assisted Medication Reconciliation in Viet Nam: Medication History Verification, Interprofessional Discrepancy Resolution and Safe Care Transitions

Governing AI-Assisted Medication Reconciliation in Viet Nam: Medication History Verification, Interprofessional Discrepancy Resolution and Safe Care Transitions

An older patient is admitted to the hospital with pneumonia. The electronic record contains several medicines prescribed during previous outpatient visits, but it does not fully reflect what the patient has been taking at home. Some medicines were stopped weeks earlier, another was recently adjusted, and a treatment prescribed at a different healthcare facility is missing.

The patient brings several medication packages to the hospital. Some product names differ from those recorded electronically. A family member remembers which tablets are usually taken in the morning but cannot confirm their strengths or recent dose changes.

An artificial intelligence (AI) system reviews the available records and produces a consolidated medication list. It also identifies differences between the documented history and the medicines ordered on admission.

The list appears useful, but important questions remain. Which information is accurate? Does a missing medicine represent an unintended omission or a deliberate treatment decision? Has the patient actually been taking the medicine as prescribed? Who is responsible for resolving these uncertainties before they affect care?

These questions illustrate why medication reconciliation is more than comparing medication names across clinical records.

AI can help healthcare professionals retrieve, organize, and compare medication information. However, combining available records does not necessarily establish what a patient has been taking. Nor should an AI-generated comparison independently determine which medicines must be continued, withheld, changed, or discontinued.

For healthcare organizations in Viet Nam, the opportunity is to use AI within a reliable clinical process that connects verified medication histories, accountable professional decisions, and clear communication whenever patients move between care settings.

Why Medication Reconciliation Matters at Care Transitions

Medication information can change considerably as patients move through healthcare services.

A medicine prescribed during an outpatient consultation may later be adjusted by another specialist. Hospital admission may require temporary interruption of treatment because of acute illness. During intensive care, medicines may be replaced, administered through different routes, or adjusted according to organ function. At discharge, some previous medicines may need to be resumed while others are permanently discontinued.

Each transition creates an opportunity for medication information to become incomplete, outdated, or misunderstood.

The World Health Organization (WHO) identifies medication safety during transitions of care as one of three priority areas under its Medication Without Harm initiative, alongside polypharmacy and high-risk situations.

WHO's 2019 technical report, Medication Safety in Transitions of Care, emphasizes structured reconciliation processes, reliable information, workforce capability, patient and caregiver involvement, and clear organizational responsibilities.

Medication discrepancies can contribute to unintended omissions, therapeutic duplication, incorrect doses, inappropriate continuation of discontinued treatments, or failure to resume medicines that remain clinically necessary.

However, not every difference between medication lists represents a medication error.

A clinician may deliberately withhold an anticoagulant because of bleeding risk, adjust a dose because of impaired kidney function, or temporarily interrupt treatment during an acute illness. These decisions may be appropriate. The safety concern arises when a difference is unintended, the reason for a change is unclear, or necessary decisions are not communicated and implemented.

Medication reconciliation is intended to reduce these risks by establishing an accurate understanding of medication use, comparing that information with current treatment orders, identifying clinically relevant discrepancies, and ensuring that appropriate decisions are documented and communicated.

This distinction matters when evaluating AI. A system that produces a more complete-looking medication list has not necessarily made care safer. The information still needs to be verified and interpreted in the context of the patient's condition and treatment plan.

Medication History, Reconciliation, and Medication Review

Medication history taking, medication reconciliation, and medication review are closely related activities, but they serve different purposes.

Medication history taking establishes which medicines a patient was actually using before a particular clinical encounter or care transition. This includes prescribed medicines and other relevant products, together with information about how they were taken.

Medication reconciliation compares that history with current or proposed medication orders to identify discrepancies, clarify whether they reflect intentional decisions, and ensure that necessary actions are documented and implemented.

Medication review is broader. It assesses whether the medication regimen remains appropriate, effective, and safe, considering the patient's conditions, treatment goals, potential interactions, adverse effects, and other relevant circumstances.

These activities may occur together, but they should not be treated as interchangeable.

For example, reconciliation may identify that a long-term medicine was unintentionally omitted from admission orders. A broader medication review may then determine whether the medicine remains clinically indicated or whether the treatment regimen should be changed.

The distinction is especially important when describing AI capabilities.

A system that extracts medicine names from clinical notes does not automatically establish a verified medication history. A tool that compares two medication lists may identify differences without understanding why they exist. Neither function, by itself, constitutes a comprehensive medication review.

Healthcare organizations should therefore define the intended clinical task before evaluating an AI product. Claims about its capabilities should reflect the functions that have actually been tested rather than implying that the entire reconciliation process can be automated.

Establishing a Trustworthy Medication History

A best possible medication history (BPMH) provides the foundation for effective medication reconciliation.

Its purpose is to establish the most accurate available account of the medicines a patient was actually taking immediately before the relevant care transition.

WHO's medication reconciliation approach emphasizes obtaining information through a systematic interview with the patient or an appropriate caregiver and checking it against at least one additional reliable source whenever possible.

Previous prescriptions, dispensing records, referral documents, medication packages, earlier discharge summaries, outpatient records, and information from other treating professionals can all be useful.

However, none of these sources should automatically be considered definitive.

A prescription shows what was ordered, but not necessarily what the patient obtained or took. A dispensing record indicates that a medicine was supplied, but it may not establish whether the patient used it correctly. An older discharge summary may accurately describe the treatment plan at discharge while no longer reflecting current medication use.

Information from patients and caregivers is essential but may also be incomplete. Some patients recognize tablets by appearance rather than by name. Others may be uncertain about medication strengths, dosing schedules, or changes made during recent consultations.

These difficulties become more pronounced when patients receive care from several healthcare facilities or obtain medicines from different pharmacies.

A reliable medication history should consider prescribed medicines, medicines obtained without a prescription, relevant traditional or herbal products, supplements, and medicines taken intermittently or when required.

The information needs sufficient detail for clinical use. Depending on the medicine, this may include its active ingredient, product name, strength, formulation, dose, route, frequency, timing, indication, and current treatment status.

Recent changes deserve particular attention. A medicine discontinued several weeks earlier should not automatically appear as current treatment simply because it remains in an electronic record.

Where information cannot be confirmed, the uncertainty should be documented rather than concealed.

For example, a medicine identified from an older prescription may be recorded as a possible previous treatment requiring verification. It should not become an active medication order merely because AI has extracted it.

The essential safeguard is to distinguish information that has been found from information that has been verified and information that remains uncertain.

Where AI Can Support Medication History Verification

AI-assisted medication reconciliation may involve different technologies and functions.

Natural language processing can identify medication-related information in clinical notes, referral letters, and discharge summaries. Image-processing systems may extract information from photographs of prescriptions, medication labels, or packaging. Other applications may help match brand names with generic ingredients, organize fragmented records, or identify potential differences between medication lists.

Large language models may also convert narrative information into structured medication histories.

These functions could reduce the time healthcare professionals spend searching through records, particularly when information is distributed across several documents or healthcare encounters.

However, errors in extraction or interpretation can change the clinical meaning of medication information.

A system may identify a medicine prescribed years earlier and incorrectly classify it as current treatment. It may overlook a discontinuation instruction, confuse product strength with the prescribed dose, or interpret a historical treatment plan as an active order.

Brand names and formulations create additional challenges. Different products may contain the same active ingredient, while similarly named products may have different strengths, formulations, or release characteristics.

Two records containing different brand names do not necessarily indicate two separate medicines. Conversely, matching products solely by active ingredient may conceal an important difference between immediate-release and modified-release formulations.

Medication timing introduces further risk. A medicine taken once daily should not be confused with one taken once weekly. A completed short course should not be treated as ongoing therapy, and a temporary interruption should not automatically be interpreted as permanent discontinuation.

Vietnamese clinical records may contain Vietnamese and English terminology, handwritten instructions, locally used abbreviations, and product names documented inconsistently across healthcare facilities.

An AI system developed using standardized clinical records from another country cannot be assumed to interpret these variations reliably.

For early implementation, AI may be most useful in retrieving and organizing medication information for professional verification rather than independently producing the definitive medication history.

Healthcare professionals should be able to identify the source of each AI-generated entry. Conflicting information should remain visible, and unverified data should not automatically overwrite existing records or trigger medication orders.

From Identifying Discrepancies to Resolving Them

The distinction between detecting a medication discrepancy and resolving it is central to safe medication reconciliation.

A discrepancy is a difference between relevant medication information sources, such as the patient's verified pre-admission history and the medicines ordered during hospitalization.

In clinical reconciliation, discrepancies are commonly classified as intentional, undocumented intentional, or unintentional.

An intentional discrepancy results from a conscious clinical decision to change treatment. For example, the admitting clinician may deliberately withhold a medicine because of the patient's acute condition.

An undocumented intentional discrepancy occurs when the change was deliberate but the decision or its rationale was not adequately recorded.

An unintentional discrepancy occurs when a difference was not the result of an intended treatment decision. An essential medicine inadvertently omitted from admission orders is one example.

These distinctions have important implications for care.

An AI system may identify that an antihypertensive medicine appears in the pre-admission history but is absent from admission orders. That difference alone does not establish a medication error. The medicine may have been intentionally withheld because of hypotension, or the patient may already have stopped taking it.

The responsible healthcare professional needs to establish what happened and determine whether corrective action is necessary.

Resolution may involve restoring an unintentionally omitted medicine, confirming that an existing order is appropriate, modifying the regimen, or documenting the reason for an intentional change.

Acknowledging an electronic alert is not the same as resolving a discrepancy.

Resolution requires an appropriate professional determination, documentation of the decision, and implementation of any necessary changes by authorized personnel. If an issue cannot be resolved before the next care transition, the remaining uncertainty and responsibility for addressing it should be communicated.

Clinical urgency also differs between discrepancies.

An unexplained interruption of an essential, time-sensitive medicine may require immediate review. A discrepancy involving a historical medicine that is no longer relevant may be addressed through routine verification.

High-alert medicines warrant particular attention because errors involving them can have serious consequences. The Institute for Safe Medication Practices identifies categories including insulin, anticoagulants, and opioids among medicines requiring additional safeguards.

Other treatments, including certain antiseizure medicines and immunosuppressants, may also create substantial risks when unintentionally interrupted or incorrectly resumed.

The appropriate response depends on the medicine, the patient's condition, and applicable clinical guidance.

AI may help highlight potentially important discrepancies or assist with prioritization. It should not independently assume that every flagged difference represents an error or automatically modify active medication orders.

Interprofessional Responsibility and Clinical Accountability

Medication reconciliation depends on contributions from several healthcare professionals, but their responsibilities are not identical.

Clinical pharmacists may help obtain and verify medication histories, identify discrepancies, assess medication-related risks, and communicate recommendations to prescribers.

Physicians and other legally authorized prescribers make medication-order decisions within their professional scope, including determining whether treatment should be continued, adjusted, withheld, or discontinued.

Nurses contribute information about medicines administered, patient responses, administration difficulties, and inconsistencies identified during routine care.

Patients and caregivers provide knowledge that may not appear in institutional records, particularly about actual medication use, recent changes, and difficulties following treatment instructions.

These contributions are complementary, but effective collaboration requires more than general agreement that medication safety is everyone's responsibility.

Healthcare organizations should establish who initiates reconciliation, who verifies the medication history, who reviews identified discrepancies, who makes treatment decisions, and who confirms that resulting orders and records have been updated.

The arrangement may differ between departments and according to available clinical pharmacy resources.

In some settings, pharmacists may conduct the initial medication history interview. Elsewhere, trained nursing staff or other appropriately qualified personnel may undertake parts of the process, with pharmacy and medical review when necessary.

Whatever model is adopted, responsibilities should be explicitly defined.

Interprofessional collaboration does not mean that all staff members have identical authority. A pharmacist's recommendation, a nurse's observation, and a prescriber's treatment decision may all be needed, but each must remain within the applicable professional scope and organizational procedures.

AI does not change this allocation of responsibility.

An automated discrepancy alert is information requiring consideration. It is not a medication order, an authorized clinical decision, or evidence that a medication-related problem has been resolved.

Medication Reconciliation at Admission, Transfer, and Discharge

Medication reconciliation should be performed at relevant care transitions rather than treated as a task completed only at hospital admission.

At admission, the immediate concern is whether the medicines ordered for the patient appropriately reflect the verified pre-admission medication history and current clinical needs.

Timing should reflect the potential consequences of errors. Patients receiving essential or time-sensitive therapies may require prompt verification of specific medicines before the complete history can be established.

The National Institute for Health and Care Excellence (NICE) guideline NG5 recommends completing medication reconciliation within 24 hours, or sooner when clinically necessary, when a person moves into an acute care setting, such as admission to hospital.

This recommendation provides an international reference for service design. It should not be interpreted as a universal statutory deadline under Vietnamese law.

Where the full medication history cannot be obtained immediately, the healthcare organization should have a process for prioritizing urgent medicines, recording unresolved information, and completing verification when reliable sources become available.

Internal transfers introduce different challenges.

A patient moving from intensive care to a general ward may have medicines that were temporarily withheld, replaced, or administered through different routes. The receiving team needs to understand which treatment decisions remain valid and whether medicines previously interrupted require reassessment.

AI may help compare medication records from successive stages of hospitalization. However, any differences must be interpreted in the context of the patient's condition and previous clinical decisions.

Discharge creates another important transition.

The discharge medication plan should distinguish medicines to continue, start, modify, or stop. Temporary interruptions should be reconsidered where appropriate, but medicines previously withheld should not be resumed automatically.

Patients need understandable information about their current regimen and clinically important changes. Where appropriate, the next healthcare provider should receive accurate medication information and the reasons for relevant treatment decisions.

An electronic discharge prescription alone does not establish that all earlier discrepancies have been resolved.

For patients transferring to another hospital, returning to outpatient care, or receiving treatment from several providers, unresolved medication questions need appropriate communication and assignment of responsibility.

A completed reconciliation process should provide a clinically reviewed medication plan, with any significant remaining uncertainties clearly documented and communicated to the professionals responsible for subsequent care.

What Current Research Tells Us

Research into AI-assisted medication reconciliation is developing, but the available evidence needs to be interpreted according to the specific functions studied.

A scoping review by Tabja Bortesi and colleagues, published in the Journal of Medical Internet Research in May 2026, examined how AI had been applied to medication reconciliation.

The review included 94 studies. All addressed subtasks related to establishing a best possible medication history, and 92 focused on automated information acquisition.

Only two studies also examined the identification of discrepancies between medication lists. None of the included studies demonstrated automation of the complete discrepancy-resolution process.

Most research concerned model development rather than routine clinical implementation. Only one included study specifically evaluated usability.

These findings illustrate an important distinction between extracting medication information and completing medication reconciliation.

The review also has a limitation relevant to its interpretation in 2026. The literature searches were conducted in June 2024, so the findings do not represent an exhaustive account of every AI application developed or evaluated afterward.

More recent studies provide examples of how AI may assist different parts of the workflow.

In February 2026, Menz and colleagues published a study in npj Digital Medicine evaluating a vision-enabled AI system for documenting medication histories.

Ten clinical pharmacists conducted 110 simulated medication history interviews. Following iterative development using ten recordings, the system was evaluated on 100 test recordings covering 2,160 data fields.

The vision-enabled system correctly documented 98% of the evaluated fields when video and audio information were available, compared with 81% when only audio was used.

The improvement was largely related to fewer omissions of information visible during the interviews, including details from medication packaging and labels.

This finding is relevant because patients frequently show medication containers, prescriptions, or tablets when explaining their treatment. A system relying only on spoken information may miss details that are clinically important but not explicitly mentioned.

However, the study involved simulated interviews. High documentation accuracy under these conditions does not establish that the system can independently verify actual medication use or safely resolve discrepancies during routine patient care.

The study also illustrates a separate governance consideration. Recording visual and audio information during clinical encounters may involve identifiable patient information and should be evaluated under applicable privacy, consent, confidentiality, and information-security requirements.

Another 2026 study by Maathuis and colleagues, published in the International Journal of Clinical Pharmacy, evaluated an AI-powered hospital admission prediction dashboard intended to support medication reconciliation prioritization in a Dutch emergency department.

The retrospective before-and-after study included 25,505 emergency department visits.

Following implementation, the proportion of patients subsequently admitted to hospital who had received medication reconciliation in the emergency department increased from 86.4% to 89.0%.

The proportion of patients discharged directly from the emergency department after receiving reconciliation decreased from 17.9% to 12.6%. The researchers classified these reconciliations as potentially unnecessary under the study's workflow definition.

This classification should be interpreted carefully. Some patients discharged from an emergency department may still require medication reconciliation because of their clinical risks, medication changes, or other circumstances. Discharge without hospital admission does not automatically mean reconciliation is unnecessary.

The findings suggest that predictive AI may help teams direct limited reconciliation resources toward patients more likely to require admission. However, the dashboard supported patient prioritization rather than performing medication history verification or discrepancy resolution.

The study did not establish whether the intervention reduced adverse drug events, hospital readmissions, or other patient harms.

A correction published in February 2026 added the study's impact statements without changing its reported results.

Together, these studies illustrate different opportunities for AI, including extracting medication information, documenting interviews, and supporting workflow prioritization.

They also demonstrate why technical performance and clinical outcomes must be distinguished.

Better information extraction does not necessarily mean a more accurate final medication history. Improved documentation does not establish that discrepancies were resolved correctly. More efficient prioritization does not, by itself, demonstrate fewer medication-related adverse events.

The United States Agency for Healthcare Research and Quality has similarly noted that evidence for medication reconciliation interventions is mixed. Some approaches improve discrepancies and process measures, while effects on readmissions and other patient outcomes are less consistent.

For healthcare organizations, AI-assisted reconciliation should therefore be assessed as part of a complete clinical service rather than as an isolated information-processing application.

The Practical Context in Vietnamese Healthcare

The feasibility of AI-assisted medication reconciliation depends on the availability of clinical pharmacy services, reliable health information systems, and established medication management processes.

A national survey of clinical pharmacy activities in Vietnamese hospitals, conducted in 2017–2018 and published by Dong and colleagues in BMC Health Services Research in 2022, found considerable variation in patient-specific clinical pharmacy activities.

Activities involving direct patient care, including obtaining medication histories, were implemented less consistently than some broader institutional pharmacy activities.

These findings describe the survey period and should not be treated as a measurement of every Vietnamese hospital in 2026. Nevertheless, they illustrate the importance of workforce capacity, clearly assigned responsibilities, and consistent clinical pharmacy processes.

Healthcare organizations in Viet Nam may differ substantially in how medication information is recorded and shared.

Some hospitals have electronic health records integrated with prescribing and pharmacy systems. Others use combinations of digital records, scanned documents, handwritten prescriptions, and patient-provided information.

Medication histories may also remain fragmented when patients receive care from several independent providers.

An AI application connected only to one hospital's records cannot reliably reconstruct a complete medication history if important information is held elsewhere.

Similarly, inconsistent medication naming, incomplete documentation, and the absence of standardized data may undermine automated matching.

These difficulties cannot necessarily be resolved by deploying a more powerful language model. They may require better information sources, clearer documentation, standardized medication terminology, and stronger coordination among healthcare professionals.

For smaller facilities or organizations with limited clinical pharmacy resources, improving a standardized manual reconciliation process may provide greater immediate value than introducing a complex AI system.

The decision to adopt AI should follow an assessment of the actual weaknesses in medication history verification, discrepancy resolution, and communication.

Local Validation Before Routine Clinical Use

Before deploying AI-assisted medication reconciliation, a healthcare organization should define which tasks the technology is expected to support.

An initial use case might involve extracting medication information from previous discharge summaries to prepare a draft history for pharmacist review.

Another implementation might identify differences between a verified pre-admission medication history and current admission orders. A third might support prioritization of patients requiring timely reconciliation.

These functions should be evaluated separately because they involve different sources of error and clinical consequences.

A system that accurately identifies medication names may still perform poorly when interpreting doses, strengths, routes, frequencies, formulations, or treatment status.

Local evaluation should therefore examine clinically relevant information fields rather than relying on a single overall accuracy figure.

The organization should test how the application handles incomplete records, conflicting information, brand–generic relationships, different formulations, abbreviations, and medication details documented in Vietnamese or English.

High-alert medicines and complex treatment regimens deserve particular attention.

A suitable reference standard should be established through qualified professional review, with a defined process for resolving disagreements among reviewers.

Evaluation should include both incorrect AI-generated information and clinically important omissions.

A high overall accuracy rate can conceal significant risks if the system performs well on common medicines but occasionally misses a discrepancy with potentially serious consequences.

Where possible, errors should be categorized according to their clinical significance.

Source traceability is equally important.

Healthcare professionals should be able to identify whether a medication entry came from a recent prescription, an older discharge summary, a dispensing record, a patient interview, or another document.

The organization should also verify that medication information is linked to the correct patient. Errors in patient identification or record matching can cause harm even when the extracted medication details are technically accurate.

Conflicting information should remain visible until verification is complete. The system should not convert uncertain information into an apparently definitive medication history.

Workflow reliability requires separate assessment.

An AI tool may reduce the time needed to prepare medication lists while increasing the time pharmacists spend correcting errors. It may generate duplicate entries, produce excessive alerts, or require repeated manual confirmation.

Hospitals should therefore evaluate technical performance alongside professional workload, ease of review, and the reliability of clinical decisions.

During supervised implementation, AI-generated information should remain subject to established verification and prescribing procedures.

The system should not independently modify active medication orders or mark clinically significant discrepancies as resolved.

Software updates, changes in medication terminology, additional information sources, or modifications to system integration may affect performance. Organizations should determine when additional testing is necessary and how significant errors will be investigated.

A successful retrospective evaluation should not automatically be considered sufficient evidence for unrestricted clinical deployment.

Measuring Whether Medication Reconciliation Improves Care

The success of AI-assisted medication reconciliation cannot be established simply by counting the number of medication lists generated or discrepancies identified.

A meaningful evaluation should determine whether medication information becomes more reliable and whether clinically important differences are addressed appropriately.

One measure is the proportion of eligible patients receiving reconciliation within a clinically appropriate timeframe.

However, completion must be defined carefully. An AI-generated medication list should not count as completed reconciliation unless the required verification, comparison, and professional review have occurred.

Organizations should also examine how often clinically significant discrepancies are identified, how many are resolved, and whether urgent issues are addressed before they can affect treatment.

Detection and resolution should be measured separately.

A system that produces many alerts may appear effective, but some differences may concern historical medicines, duplicate entries, or intentional treatment changes that require no correction.

Excessive low-value alerts can increase workload and make important discrepancies harder to recognize.

Useful measures include the reliability of medication histories, the proportion of clinically significant discrepancies resolved, the time required to address urgent issues, and the completeness of medication information communicated during transfer or discharge.

Balancing measures are also necessary.

These may include incorrect AI-generated entries, inappropriate treatment changes, duplicated documentation, unnecessary interruptions to clinical work, and additional time spent correcting system outputs.

Staff workload and implementation costs should also be considered, particularly where clinical pharmacy resources are limited.

Where reliable information is available, longer-term evaluation may include medication errors, adverse drug events, medication-related emergency visits, and hospital readmissions.

These outcomes require careful interpretation.

More complete incident reporting may increase the number of documented medication errors without indicating that care has become less safe. Conversely, fewer recorded discrepancies may reflect poorer detection rather than a genuine improvement.

A stronger evaluation compares the assisted process with an appropriate baseline or comparison group and considers changes in staffing, patient complexity, documentation practices, and clinical services.

The central question is whether the complete intervention, including AI support, professional verification, discrepancy resolution, and communication, provides safer and more reliable medication management than the previous process.

Legal, Professional, and Data Governance Considerations in Viet Nam

AI-assisted medication reconciliation must operate within Viet Nam's legal framework for healthcare practice, clinical pharmacy, medication management, artificial intelligence, medical devices, and personal data protection.

These frameworks address different responsibilities. Compliance with one does not automatically establish compliance with another or demonstrate that a clinical application is safe.

Clinical Practice and Pharmacy Responsibilities

The Law on Medical Examination and Treatment No. 15/2023/QH15, effective January 1, 2024, as subsequently amended, provides an important legal foundation for professional healthcare practice, patient rights, clinical records, and organizational responsibilities.

Decree No. 131/2020/NĐ-CP, effective January 1, 2021, regulates the organization and activities of clinical pharmacy services in healthcare facilities.

Ministry of Health Circular No. 23/2011/TT-BYT, effective July 25, 2011, remains relevant to medication use in healthcare facilities with inpatient beds, subject to applicable amendments and related provisions.

For outpatient prescriptions involving chemical drugs and biological products, Circular No. 26/2025/TT-BYT, effective July 1, 2025, provides additional requirements concerning prescriptions and prescribing.

These instruments form part of the professional and operational context in which medication reconciliation occurs. They should not be interpreted as establishing a universal mandatory AI-assisted medication reconciliation model.

Healthcare organizations need to determine how AI-supported activities fit within existing responsibilities for medication history taking, clinical pharmacy review, prescribing, medication administration, and clinical documentation.

The use of an AI application does not independently transfer prescribing authority or other regulated professional responsibilities to the software.

Artificial Intelligence Regulation

Viet Nam's Law on Artificial Intelligence No. 134/2025/QH15 took effect on March 1, 2026.

Decree No. 142/2026/NĐ-CP, effective May 1, 2026, provides implementing provisions, including those concerning AI risk classification and responsibilities associated with AI systems.

Decision No. 33/2026/QĐ-TTg, effective August 15, 2026, establishes the List of High-Risk Artificial Intelligence Systems under the applicable framework.

An AI-assisted medication reconciliation application should not automatically be classified as high-risk or low-risk simply because it is used in healthcare.

Its legal classification depends on its intended purpose, functions, degree of automation, actual deployment, and relevant statutory criteria.

A system that extracts medication information for professional verification may present different risks from one that independently recommends treatment changes or updates active medication orders.

Healthcare organizations should therefore examine the specific product and intended workflow before determining which regulatory obligations apply.

The clinical safety assessment should also consider the potential consequences of incomplete or incorrect medication information, regardless of the system's statutory classification.

A system not expressly included in a particular high-risk category may still require substantial clinical safeguards.

Medical Device Requirements

Some AI applications may fall within Viet Nam's medical device regulatory framework, depending on their intended medical purpose and functions.

Decree No. 98/2021/NĐ-CP on medical device management, as amended, including by Decree No. 07/2023/NĐ-CP and Decree No. 04/2025/NĐ-CP, provides part of the applicable framework.

Ministry of Health Circular No. 24/2026/TT-BYT, effective July 1, 2026, addresses risk determination and management measures for products classified as medical devices and amends relevant implementing provisions.

The presence of AI does not automatically make software a medical device.

Conversely, describing an application as an administrative or workflow tool does not necessarily exclude medical device requirements if its intended functions have a medical purpose.

Product classification and applicable obligations should be determined for the specific technology and deployment context.

Regulatory authorization or certification obtained in another jurisdiction may inform product assessment but does not independently establish compliance with Vietnamese requirements.

Personal Data Protection and Information Governance

Medication reconciliation may involve identifiable diagnoses, prescription histories, dispensing information, clinical observations, medication allergies, and details of actual medication use.

The Law on Personal Data Protection No. 91/2025/QH15 and Decree No. 356/2025/NĐ-CP, both effective January 1, 2026, provide relevant requirements concerning personal data processing and protection.

Healthcare organizations should establish the applicable lawful basis and purposes for processing personal data, including relevant consent requirements or legally applicable exceptions.

Appropriate safeguards should address access permissions, information security, retention practices, record traceability, and the responsibilities of external technology providers.

Particular attention is needed when identifiable health information is processed by external AI services or transferred outside Viet Nam.

An application hosted within a hospital's own infrastructure may reduce certain external data-sharing risks. However, local deployment does not automatically establish compliance with all applicable privacy, security, and healthcare requirements.

Hospitals should understand what information the system processes, where it is stored, who can access it, and whether the technology provider may use it for purposes beyond the approved clinical activity.

Information collected to support patient care should not automatically be reused for commercial AI model training or unrelated development without assessing the applicable legal basis and required safeguards.

Clinical accountability also requires reliable record management. AI-generated medication information should be distinguishable from verified clinical information and authorized treatment decisions.

Corrections should be traceable through appropriate clinical documentation procedures.

International recommendations from WHO, NICE, and other professional organizations may inform local implementation, but they do not automatically establish binding Vietnamese legal requirements.

The governance measures described in this article are proposed approaches to responsible implementation unless a specific applicable legal or professional requirement provides otherwise.

A Practical Starting Point for Healthcare Organizations

A hospital considering AI-assisted medication reconciliation should begin by understanding how medication information currently moves through its clinical services.

A baseline review could examine recent admissions, internal transfers, and discharges to determine how medication histories are obtained, where discrepancies occur, and whether clinically important differences are resolved.

The assessment should involve clinical pharmacists, physicians, nurses, information technology personnel, and healthcare quality teams.

It may reveal that the main problem is incomplete medication history taking, limited access to external records, unclear responsibility for resolving discrepancies, or poor communication between departments.

Some of these weaknesses may be addressed more effectively through clearer documentation, staff training, and standardized reconciliation procedures than through AI.

If AI is considered appropriate, the organization should select a limited initial use case.

For example, a hospital may evaluate whether AI-assisted extraction reduces the time required to prepare draft medication histories from existing records without compromising clinical accuracy.

The system should initially support professional verification rather than independently alter active medication orders.

A supervised evaluation can compare AI-generated information with medication histories established by qualified healthcare professionals. Clinically important errors, missing medicines, incorrect doses, and problems with source interpretation should be investigated.

The organization can then assess whether the assisted process improves history completeness, discrepancy identification, resolution time, and staff workload.

Where results support further use, a controlled pilot may evaluate performance during actual admission or discharge workflows.

The pilot should retain established prescribing and verification responsibilities, define escalation procedures, and include arrangements for responding to incorrect AI outputs and system failures.

Patient information used during testing and implementation must be handled under applicable legal and organizational safeguards.

As experience develops, AI support may be extended to additional departments or care transitions.

Expansion should depend on demonstrated clinical reliability, acceptable risk, and practical benefit rather than the number of records processed.

A dependable medication reconciliation process must also continue when AI is unavailable. Essential medication verification, prescribing decisions, and communication must not depend entirely on the technology.

From Better Medication Information to Safer Care Transitions

Medication reconciliation illustrates an important distinction between processing clinical information and delivering safe care.

AI may become increasingly capable of extracting medication histories, identifying inconsistencies, and organizing information for professional review. These functions can be valuable when records are fragmented and healthcare professionals have limited time to verify them.

However, the most consequential questions remain clinical.

A medicine missing from an admission order may represent an unintended omission or a deliberate treatment decision. A medicine appearing in an older record may no longer be used. A treatment withheld during hospitalization may need reassessment before discharge.

Understanding these differences requires reliable information, professional judgment, and communication among the people responsible for the patient's care.

For healthcare organizations in Viet Nam, responsible AI-assisted medication reconciliation should connect three essential activities: establishing what the patient has actually been taking, resolving clinically relevant discrepancies through accountable professional decisions, and communicating an accurate medication plan whenever care changes hands.

The value of AI should not be measured by how many medicine names it extracts or how many differences it flags.

It should be assessed by whether clinically important uncertainties are resolved, preventable medication errors are addressed, and patients and healthcare professionals receive the information needed to continue treatment safely.

AI can help organize medication information. Safe reconciliation depends on what healthcare teams do with that information.

Selected References and Regulatory Sources

International Patient-Safety Guidance

World Health Organization. Medication Safety in Transitions of Care. Geneva: WHO; 2019. WHO/UHC/SDS/2019.9.

World Health Organization. The High5s Project: Standard Operating Protocol for Medication Reconciliation. Version 3, September 2014. WHO publication record, 2021.

National Institute for Health and Care Excellence. Medicines Optimisation: The Safe and Effective Use of Medicines to Enable the Best Possible Outcomes. NICE Guideline NG5. 2015, as updated.

Agency for Healthcare Research and Quality. Medications at Transitions and Clinical Handoffs (MATCH) Toolkit for Medication Reconciliation.

Institute for Safe Medication Practices. ISMP List of High-Alert Medications in Acute Care Settings. 2024.

AI and Medication Reconciliation Research

Tabja Bortesi JP, Becerra MP, Ranisau J, et al. AI-Based Automation for Medication Reconciliation: Scoping Review. Journal of Medical Internet Research. 2026;28. DOI: 10.2196/86760.

Menz BD, Scarfo NL, Modi ND, et al. Vision-Enabled AI Scribes Reduce Omissions in Clinical Conversations: Evidence From Simulated Medication Histories. npj Digital Medicine. 2026;9:287. DOI: 10.1038/s41746-026-02494-9.

Maathuis J, Veldhuis A, Egbers JB, et al. Impact of an AI-Powered Hospital Admission Prediction Dashboard to Guide Medication Reconciliation in the Emergency Department: A Retrospective Before-After Study. International Journal of Clinical Pharmacy. 2026;48:801–807. DOI: 10.1007/s11096-025-02076-x. Correction published February 2026; DOI: 10.1007/s11096-026-02100-8.

Agency for Healthcare Research and Quality. Medication Reconciliation. PSNet Patient Safety Primer. Reviewed December 2024.

Vietnamese Healthcare Implementation Evidence

Dong PTX, Trinh HT, Nguyen DH, et al. Implementing Clinical Pharmacy Activities in Hospital Setting in Vietnam: Current Status From a National Survey. BMC Health Services Research. 2022;22:878. DOI: 10.1186/s12913-022-08242-5.

Relevant Vietnamese Legal and Regulatory Framework

National Assembly of Viet Nam. Law on Medical Examination and Treatment No. 15/2023/QH15, effective January 1, 2024, as subsequently amended.

Government of Viet Nam. Decree No. 131/2020/NĐ-CP on the Organization and Activities of Clinical Pharmacy in Healthcare Facilities, effective January 1, 2021.

Ministry of Health of Viet Nam. Circular No. 23/2011/TT-BYT on Medication Use in Healthcare Facilities With Inpatient Beds, effective July 25, 2011, subject to applicable amendments.

Ministry of Health of Viet Nam. Circular No. 26/2025/TT-BYT on Prescriptions and Prescribing Chemical Drugs and Biological Products in Outpatient Treatment, effective July 1, 2025.

National Assembly of Viet Nam. Law on Artificial Intelligence No. 134/2025/QH15, effective March 1, 2026.

Government of Viet Nam. Decree No. 142/2026/NĐ-CP, detailing certain provisions and measures for implementing the Law on Artificial Intelligence, effective May 1, 2026.

Prime Minister of Viet Nam. Decision No. 33/2026/QĐ-TTg, promulgating the List of High-Risk Artificial Intelligence Systems, effective August 15, 2026.

Government of Viet Nam. Decree No. 98/2021/NĐ-CP on Medical Device Management, as amended, including by Decree No. 07/2023/NĐ-CP and Decree No. 04/2025/NĐ-CP.

Ministry of Health of Viet Nam. Circular No. 24/2026/TT-BYT on Risk Determination and Management Measures for Medical Devices and Amendments to Related Implementing Provisions, effective July 1, 2026.

National Assembly of Viet Nam. Law on Personal Data Protection No. 91/2025/QH15, effective January 1, 2026.

Government of Viet Nam. Decree No. 356/2025/NĐ-CP, detailing certain provisions and measures for implementing the Law on Personal Data Protection, effective January 1, 2026.

Editorial and Clinical Notice

This article is intended for professional education, responsible healthcare AI governance, and healthcare quality-improvement discussion. The opening patient scenario is illustrative and does not describe an identified individual or documented clinical incident.

The article summarizes selected international patient-safety guidance, published research, and relevant Vietnamese legal and regulatory considerations. It also presents proposed organizational approaches for medication history verification, interprofessional discrepancy resolution, safe care transitions, and performance monitoring.

These approaches do not constitute a mandatory Vietnamese AI implementation standard, an independently validated medication reconciliation protocol, or a legal determination of any specific AI product's regulatory classification.

The article does not provide patient-specific medication advice, authorize automated prescribing or medication changes, or replace professional clinical judgment.

Medication history verification, discrepancy resolution, prescribing decisions, and communication across care transitions must remain under the responsibility of appropriately qualified healthcare professionals, consistent with applicable Vietnamese requirements and locally approved procedures.

The evidence discussed does not establish that AI-assisted medication reconciliation necessarily reduces adverse drug events, hospital readmissions, or other patient harms in Vietnamese healthcare settings. Local evaluation, supervised implementation, and ongoing monitoring remain necessary before claims of clinical effectiveness can be made.

Digital Medicine Vietnam (DMVN) | Evidence and regulatory review: 1 November 2026.